[Bug]: vLLM ModelConfig doesn't pass hf_overrides to get_hf_image_processor_config, which could contain auth token for hugging face (not in ENV)

[void-mckenzie]

Your current environment

The output of `python collect_env.py`

INFO 03-14 23:03:57 __init__.py:183] Automatically detected platform cuda.
Collecting environment information...
PyTorch version: 2.5.1+cu124
Is debug build: False
CUDA used to build PyTorch: 12.4
ROCM used to build PyTorch: N/A

OS: Linux Mint 21.3 (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: version 3.22.1
Libc version: glibc-2.35

Python version: 3.12.8 | packaged by Anaconda, Inc. | (main, Dec 11 2024, 16:31:09) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-6.8.0-52-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.4.131
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: GPU 0: NVIDIA GeForce RTX 4090
Nvidia driver version: 550.144.03
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture:                         x86_64
CPU op-mode(s):                       32-bit, 64-bit
Address sizes:                        48 bits physical, 48 bits virtual
Byte Order:                           Little Endian
CPU(s):                               32
On-line CPU(s) list:                  0-31
Vendor ID:                            AuthenticAMD
Model name:                           AMD Ryzen 9 7950X3D 16-Core Processor
CPU family:                           25
Model:                                97
Thread(s) per core:                   2
Core(s) per socket:                   16
Socket(s):                            1
Stepping:                             2
CPU max MHz:                          5858.0000
CPU min MHz:                          545.0000
BogoMIPS:                             8399.98
Flags:                                fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local user_shstk avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid overflow_recov succor smca fsrm flush_l1d
Virtualization:                       AMD-V
L1d cache:                            512 KiB (16 instances)
L1i cache:                            512 KiB (16 instances)
L2 cache:                             16 MiB (16 instances)
L3 cache:                             128 MiB (2 instances)
NUMA node(s):                         1
NUMA node0 CPU(s):                    0-31
Vulnerability Gather data sampling:   Not affected
Vulnerability Itlb multihit:          Not affected
Vulnerability L1tf:                   Not affected
Vulnerability Mds:                    Not affected
Vulnerability Meltdown:               Not affected
Vulnerability Mmio stale data:        Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed:               Not affected
Vulnerability Spec rstack overflow:   Mitigation; Safe RET
Vulnerability Spec store bypass:      Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:             Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:             Mitigation; Enhanced / Automatic IBRS; IBPB conditional; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Not affected

Versions of relevant libraries:
[pip3] flake8==7.1.1
[pip3] mypy-extensions==1.0.0
[pip3] numpy==1.26.4
[pip3] numpydoc==1.7.0
[pip3] nvidia-cublas-cu12==12.4.5.8
[pip3] nvidia-cuda-cupti-cu12==12.4.127
[pip3] nvidia-cuda-nvrtc-cu12==12.4.127
[pip3] nvidia-cuda-runtime-cu12==12.4.127
[pip3] nvidia-cudnn-cu12==9.1.0.70
[pip3] nvidia-cufft-cu12==11.2.1.3
[pip3] nvidia-curand-cu12==10.3.5.147
[pip3] nvidia-cusolver-cu12==11.6.1.9
[pip3] nvidia-cusparse-cu12==12.3.1.170
[pip3] nvidia-cusparselt-cu12==0.6.2
[pip3] nvidia-ml-py==12.570.86
[pip3] nvidia-nccl-cu12==2.21.5
[pip3] nvidia-nvjitlink-cu12==12.4.127
[pip3] nvidia-nvtx-cu12==12.4.127
[pip3] onnxruntime==1.20.1
[pip3] pyzmq==26.2.0
[pip3] rapidocr-onnxruntime==1.3.24
[pip3] sentence-transformers==3.3.1
[pip3] torch==2.5.1
[pip3] torchvision==0.20.1
[pip3] transformers==4.46.3
[pip3] triton==3.1.0
[conda] numpy                     1.26.4                   pypi_0    pypi
[conda] numpydoc                  1.7.0           py312h06a4308_0  
[conda] nvidia-cublas-cu12        12.4.5.8                 pypi_0    pypi
[conda] nvidia-cuda-cupti-cu12    12.4.127                 pypi_0    pypi
[conda] nvidia-cuda-nvrtc-cu12    12.4.127                 pypi_0    pypi
[conda] nvidia-cuda-runtime-cu12  12.4.127                 pypi_0    pypi
[conda] nvidia-cudnn-cu12         9.1.0.70                 pypi_0    pypi
[conda] nvidia-cufft-cu12         11.2.1.3                 pypi_0    pypi
[conda] nvidia-curand-cu12        10.3.5.147               pypi_0    pypi
[conda] nvidia-cusolver-cu12      11.6.1.9                 pypi_0    pypi
[conda] nvidia-cusparse-cu12      12.3.1.170               pypi_0    pypi
[conda] nvidia-cusparselt-cu12    0.6.2                    pypi_0    pypi
[conda] nvidia-ml-py              12.570.86                pypi_0    pypi
[conda] nvidia-nccl-cu12          2.21.5                   pypi_0    pypi
[conda] nvidia-nvjitlink-cu12     12.4.127                 pypi_0    pypi
[conda] nvidia-nvtx-cu12          12.4.127                 pypi_0    pypi
[conda] pyzmq                     26.2.1                   pypi_0    pypi
[conda] sentence-transformers     3.3.1                    pypi_0    pypi
[conda] torch                     2.5.1                    pypi_0    pypi
[conda] torchvision               0.20.1                   pypi_0    pypi
[conda] transformers              4.46.3                   pypi_0    pypi
[conda] triton                    3.1.0                    pypi_0    pypi
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.7.0
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0	CPU Affinity	NUMA Affinity	GPU NUMA ID
GPU0	 X 	0-31	0		N/A

Legend:

  X    = Self
  SYS  = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
  NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
  PHB  = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
  PXB  = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
  PIX  = Connection traversing at most a single PCIe bridge
  NV#  = Connection traversing a bonded set of # NVLinks

LD_LIBRARY_PATH=/home/mukmckenzie/anaconda3/envs/llm/lib/python3.12/site-packages/cv2/../../lib64:/usr/lib/x86_64-linux-gnu:
NCCL_CUMEM_ENABLE=0
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY

🐛 Describe the bug

vLLM ModelConfig in config.py doesn't pass hf_overrides to get_hf_image_processor_config.

if hf_overrides_kw:
    logger.info("Overriding HF config with %s", hf_overrides_kw)
    hf_config.update(hf_overrides_kw)
if hf_overrides_fn:
    logger.info("Overriding HF config with %s", hf_overrides_fn)
    hf_config = hf_overrides_fn(hf_config)

self.hf_config = hf_config

self.hf_text_config = get_hf_text_config(self.hf_config)
self.encoder_config = self._get_encoder_config()
self.hf_image_processor_config = get_hf_image_processor_config(
    self.model, revision)

####################

def get_hf_image_processor_config(
    model: Union[str, Path],
    revision: Optional[str] = None,
    **kwargs,
) -> Dict[str, Any]:
    # ModelScope does not provide an interface for image_processor
    if VLLM_USE_MODELSCOPE:
        return dict()
    # Separate model folder from file path for GGUF models
    if check_gguf_file(model):
        model = Path(model).parent
    return get_image_processor_config(model, revision=revision, **kwargs)

####################
#transformers/models/auto/image_processing_auto.py
def get_image_processor_config(
    pretrained_model_name_or_path: Union[str, os.PathLike],
    cache_dir: Optional[Union[str, os.PathLike]] = None,
    force_download: bool = False,
    resume_download: Optional[bool] = None,
    proxies: Optional[Dict[str, str]] = None,
    token: Optional[Union[bool, str]] = None,
    revision: Optional[str] = None,
    local_files_only: bool = False,
    **kwargs,
):
use_auth_token = kwargs.pop("use_auth_token", None)
    if use_auth_token is not None:
        warnings.warn(
            "The `use_auth_token` argument is deprecated and will be removed in v5 of Transformers. Please use `token` instead.",
            FutureWarning,
        )
        if token is not None:
            raise ValueError("`token` and `use_auth_token` are both specified. Please set only the argument `token`.")
        token = use_auth_token
......

This hf_overrides might contain the auth token from huggingface. From unsloth, the only way to pass the auth_token to vllm (not in env), is through hf_overrides. When pulling custom models from hugging face through vllm, we face this error:

RuntimeError: void-mckenzie/krikri-sft_compound_instruct is not a local folder and is not a valid model identifier listed on 'https://huggingface.co/models'
If this is a private repository, make sure to pass a token having permission to this repo either by logging in with `huggingface-cli login` or by passing `token=<your_token>`

This limits the usage of multiple HF Tokens with vLLM. As you can see from the above snippet, get_image_processor_config in transformers accepts token. We just need to pass it when calling get_hf_image_processor_config from vllm/config.py

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[DarkLight1337]

I don't think using hf_overrides is a good way to pass tokens as they will be logged as shown in the code snippet you have posted (possibly leaking the token). Is it not possible to pass the token via environment variables?

[void-mckenzie]

@DarkLight1337 It works fine if you just have one HF token. But if you have multiple, you need to reset the environment variable every time. Wouldn't it make more sense to use the inbuilt auth token functionality in transformers? It exists to support multiple token access. I have an issue right now with my unsloth build, since vllm cuts off the auth token pathway from unsloth to transformers.

[DarkLight1337]

What API is required for vLLM to support this? I'm not familiar with how this works in Unsloth.

[void-mckenzie]

Unsloth uses vLLM for fast inference. It calls a LLM class (vllm/entrypoints/llm.py) with the vLLM defined set of arguments.

This doesn't take in a token, so we need to pass the token through hf_overrides (supported by the class).

This is passed on to the create_engine_config (vllm/engine/arg_utils.py) , which passes on to ModelConfig.

In this class, the hf_override variable is properly parsed. But, it's only sent to get_hf_text_config(). This method doesn't make use of the any auth token.

The hf_overrides parameter is however, not passed to the subsequent get_hf_image_process. This method internally called get_image_processor_config(), which can take a token as a parameter. If passed through get_hf_image_process, we can achieve multiple token support.

I have tried my best to explain what happens here. Let me know if you need anything else, or if it's too confusing.

[DarkLight1337]

If Unsloth supports directly passing through the LLM arguments, we can introduce a new LLM engine argument to pass the tokens so they aren't logged as I noted previously.

[void-mckenzie]

I can set up a PR on Unsloth to do that. This would be very helpful, thank you. Let me know once this feature is supported, or if I can help in any other way. Thank you.

[DarkLight1337]

I'm busy with other PRs, so let's see if anyone is available to take this up on vLLM side